Notes on frontend architecture, design systems, trustworthy AI interfaces, developer experience, and shipping software that holds up.
These are the closest thing to watching me think: how I reason about architecture, product constraints, data integrity, performance, and where automated systems should stop and a person should decide.
Upcoming
What I changed in Agent Readiness 0.1.2 to prepare for directory review: plugin documentation, schema validation, honest failure behavior, coordinated versions, and tests of the installed artifact.
Preview article →A refactor can leave the code healthy and its instructions wrong. How I would use Agent Readiness Doctor to make instruction maintenance part of the development workflow.
Read article →A practical frontend design for reviewing AI-extracted documents: bounded queues, PDF-field selection, server-enforced leases, recoverable drafts, and accessible review workflows.
Read article →A take-home brief asked for one function that classifies sensors from a log file. How I approached it as a library: reading the spec for ambiguity, streaming with fixed state, failing loudly, making new sensor types cheap, and the memory bug I only found after submitting.
Read article →A CI check that runs on pull requests reads input written by whoever opened the PR. Lessons from hardening three small open-source tools: package names, git refs, file names, config, symlinks, and bot comments.
Read article →Agent-generated pull requests touch more files than a reviewer can read with equal care. A deterministic pre-screen decides where human attention goes first, and why I kept an LLM out of that decision.
Read article →AGENTS.md, CLAUDE.md, and Cursor rules steer every change an AI coding agent makes, but almost nobody reviews them. What goes wrong in those files, and the deterministic checks I built to catch it.
Read article →Featured
Frontend architecture is not a folder layout or a state library. It is deciding which kind of state each piece of data is, giving it exactly one owner, and making the second owner hard to create.
Read article →A practical frontend architecture approach for a chess.com-style backgammon product: real-time game state, SVG rendering, the play-improve-compete loop, accessibility, AI-assisted analysis, and the infrastructure that makes the UI trustworthy.
Read article →A practical look at how I use AI in frontend engineering across planning, implementation, review, testing, and model evaluation without outsourcing engineering judgment.
Read article →A practical view of senior frontend architecture: boundaries, state, API contracts, accessibility, performance, test strategy, and the review habits that keep product UI maintainable.
Read article →On LedgerGuard, the hard problem is not 'can we extract dates from a PDF?' It is making renewal and spend readouts stay honest when queues, idempotent workers, and humans disagree.
Read article →How I moved from vague, ChatGPT-flavored suggestions to grounded, auditable AI recommendations inside a real zoning and feasibility product.
Read article →A short framework for evaluating whether a design system, developer tool, or AI feature is actually helping the work instead of adding another layer around it.
Read article →Why I use MCP to run deterministic local checks in the editor, what that buys over scripts or hooks alone, and where human review still starts.
Read article →How to get adoption and keep a design system useful with better component contracts, practical documentation, and lightweight governance.
Read article →